欧盟 AI 监管框架体系：从 AI Act 到 CADA 的完整治理架构

# AI Act 高风险AI系统快速评估工具
# 用途：帮助企业和开发者初步判断其 AI 系统是否属于高风险类别
# 注意：这是简化版评估工具，正式合规需要法律专业人士参与

class AIActRiskClassifier:
    """
    基于 AI Act Annex I 的高风险 AI 系统分类器
    """

    HIGH_RISK_CATEGORIES = {
        "biometrics": "生物识别（远程生物识别、情绪识别）",
        "critical_infrastructure": "关键基础设施管理（水、气、热、电）",
        "education": "教育和职业培训（招生筛选、考试评分）",
        "employment": "就业和人才管理（招聘、晋升、任务分配）",
        "public_services": "公共服务（信用评估、紧急调度）",
        "law_enforcement": "执法（犯罪风险评估、测谎）",
        "migration": "移民和边境管理（签证审核、庇护评估）",
        "justice": "司法和民主程序（法律研究辅助、选举评估）",
    }

    PROHIBITED_PRACTICES = [
        "social_scoring",      # 社会评分系统
        "exploitation",         # 利用人类弱点的操纵
        "remote_biometric",    # 实时远程生物识别监控（有限例外）
        "emotion_workplace",   # 工作场所情绪识别
        "predictive_policing", # 基于个人特征的犯罪预测
    ]

    def classify_system(self, use_case: str, deployment_context: str) -> dict:
        """对 AI 系统进行初步风险分类"""
        result = {
            "use_case": use_case,
            "context": deployment_context,
            "risk_level": "unknown",
            "category": None,
            "recommendations": [],
        }

        # 1. 检查是否为禁止性实践
        if self._is_prohibited(use_case):
            result["risk_level"] = "unacceptable"
            result["recommendations"].append("该系统属于禁止性实践，不得在欧盟市场部署")
            return result

        # 2. 检查是否为高风险类别
        for cat_key, cat_desc in self.HIGH_RISK_CATEGORIES.items():
            if self._matches_category(use_case, cat_key):
                result["risk_level"] = "high"
                result["category"] = cat_desc
                result["recommendations"].extend([
                    "建立风险管理体系（AI Act 第9条）",
                    "准备技术文档（AI Act 第11条）",
                    "确保数据治理合规（AI Act 第10条）",
                    "设计人类监督机制（AI Act 第14条）",
                    "通过合格评定程序（AI Act 第43条）",
                ])
                return result

        # 3. 检查是否需要透明度义务
        if self._needs_transparency(use_case):
            result["risk_level"] = "limited"
            result["recommendations"].append("确保用户知晓正在与 AI 交互（AI Act 第50条）")
            return result

        # 4. 最小风险
        result["risk_level"] = "minimal"
        result["recommendations"].append("无额外义务，但建议遵循自愿性行为准则")
        return result

    def _is_prohibited(self, use_case: str) -> bool:
        prohibited_keywords = {
            "social_scoring": ["社会评分", "social scoring", "信用评分公民"],
            "exploitation": ["利用弱点", "exploit vulnerability", "subliminal"],
            "remote_biometric": ["远程生物识别", "remote biometric", "real-time facial"],
        }
        for practice, keywords in prohibited_keywords.items():
            if any(kw.lower() in use_case.lower() for kw in keywords):
                return True
        return False

    def _matches_category(self, use_case: str, category: str) -> bool:
        category_keywords = {
            "biometrics": ["biometric", "facial recognition", "情绪识别", "指纹"],
            "critical_infrastructure": ["电力", "water", "gas", "基础设施"],
            "education": ["admission", "exam scoring", "招生", "考试评分"],
            "employment": ["hiring", "recruitment", "招聘", "晋升"],
            "public_services": ["credit scoring", "信用评估"],
            "law_enforcement": ["crime prediction", "犯罪预测", "lie detection"],
            "migration": ["visa", "asylum", "签证", "庇护"],
            "justice": ["legal research", "election", "法律研究", "选举"],
        }
        keywords = category_keywords.get(category, [])
        return any(kw.lower() in use_case.lower() for kw in keywords)

    def _needs_transparency(self, use_case: str) -> bool:
        transparency_keywords = ["chatbot", "聊天机器人", "deepfake", "深度伪造", "AI generated"]
        return any(kw.lower() in use_case.lower() for kw in transparency_keywords)


# 使用示例
if __name__ == "__main__":
    classifier = AIActRiskClassifier()

    test_cases = [
        ("AI 招聘简历筛选系统", "企业内部 HR 使用"),
        ("聊天机器人客服", "面向欧盟消费者"),
        ("医疗影像辅助诊断", "医院放射科使用"),
    ]

    for use_case, context in test_cases:
        result = classifier.classify_system(use_case, context)
        print(f"\n用例: {use_case}")
        print(f"环境: {context}")
        print(f"风险等级: {result['risk_level']}")
        if result['category']:
            print(f"类别: {result['category']}")
        for rec in result['recommendations']:
            print(f"  - {rec}")

# AI Act 合规成本估算模型
# 用途：估算企业完成 AI Act 高风险合规的初期和持续成本

def estimate_ai_act_compliance_cost(
    company_size: str,       # "small", "medium", "large", "enterprise"
    num_ai_systems: int,     # AI 系统数量
    high_risk_count: int,    # 高风险 AI 系统数量
    has_existing_qms: bool = False,  # 是否已有质量管理体系
) -> dict:
    """
    估算 AI Act 高风险合规成本

    基于 Bird & Bird、DLA Piper 等法律机构的公开估算数据
    """
    # 基础成本（欧元）
    base_costs = {
        "risk_assessment": 10000,     # 每个系统的风险评估
        "tech_doc": 30000,            # 每个高风险系统的技术文档
        "qms_setup": 40000,           # 质量管理体系建设（一次性）
        "audit_cert": 20000,          # 第三方审计认证（每个系统）
        "annual_ops": 30000,          # 持续合规运营（每年）
    }

    # 公司规模系数
    size_multiplier = {
        "small": 0.5,
        "medium": 0.8,
        "large": 1.0,
        "enterprise": 1.5,
    }

    multiplier = size_multiplier.get(company_size, 1.0)

    # 计算各项成本
    risk_cost = base_costs["risk_assessment"] * num_ai_systems * multiplier
    doc_cost = base_costs["tech_doc"] * high_risk_count * multiplier
    qms_cost = base_costs["qms_setup"] * multiplier if not has_existing_qms else 0
    audit_cost = base_costs["audit_cert"] * high_risk_count * multiplier
    annual_cost = base_costs["annual_ops"] * multiplier

    initial_cost = risk_cost + doc_cost + qms_cost + audit_cost
    total_first_year = initial_cost + annual_cost

    return {
        "company_size": company_size,
        "initial_cost_eur": round(initial_cost),
        "annual_cost_eur": round(annual_cost),
        "total_first_year_eur": round(total_first_year),
        "breakdown": {
            "risk_assessment": round(risk_cost),
            "technical_docs": round(doc_cost),
            "qms_setup": round(qms_cost),
            "audit_certification": round(audit_cost),
            "annual_operations": round(annual_cost),
        },
    }


# 使用示例
if __name__ == "__main__":
    # 中型企业，5个AI系统，2个高风险
    cost = estimate_ai_act_compliance_cost(
        company_size="medium",
        num_ai_systems=5,
        high_risk_count=2,
        has_existing_qms=False,
    )

    print(f"=== AI Act 合规成本估算（中型企业）===")
    print(f"初期投入: {cost['initial_cost_eur']:,} 欧元")
    print(f"年度运营: {cost['annual_cost_eur']:,} 欧元/年")
    print(f"首年总计: {cost['total_first_year_eur']:,} 欧元")
    print("\n费用分解:")
    for k, v in cost['breakdown'].items():
        print(f"  {k}: {v:,} 欧元")